What Happens When Scientists Put an AI-Powered Banana Casino in a Monkey Reserve?

Researchers from Emory University and the Georgia Institute of Technology built an AI-powered touchscreen device that hands out banana slices to wild capuchin monkeys who complete a simple task, and the early results read a lot like a casino slot machine study. Dubbed CapuchinAI, the system uses facial recognition to spot an approaching monkey, offers up a task on a screen, and dispenses a reward the moment the animal gets it right.

The team published its findings in the American Journal of Primatology, describing the device as a proof-of-concept for studying primate cognition in the wild rather than in a lab. For anyone who follows how humans respond to reward loops in a casino, the parallels in monkey behavior are hard to ignore — habituation, repeat visits, and individual differences in how quickly a subject chases the payout. It’s the same basic loop that keeps players tapping a slot reel or opening a social casino app for another spin, just dressed up in banana slices instead of coins.

Building a Banana Vending Machine for the Jungle

CapuchinAI combines a small Raspberry Pi computer, a webcam, a touchscreen, and a 3D-printed food dispenser inside a weatherproofed box that runs for roughly eight hours on ordinary travel battery packs. The facial-recognition pipeline, adapted from the open-source YOLO computer vision architecture, was trained on thousands of images and video clips of six adult male capuchins living at Costa Rica’s Taboga Forest Reserve.

Once trained, the system detected a capuchin’s face in frame 98% of the time and correctly identified which individual monkey it was looking at 97% of the time, even when working from blurred or differently lit footage than what it trained on. Researchers first taught the monkeys what the box was for with a simple blue square on the screen — touch it, and the machine drops a piece of dried banana.

Some Monkeys Learned Fast, Others Just Watched

It took a few days before a dominant male capuchin approached the strange device, according to study co-author Federico Sánchez Vargas of Emory University. He initially climbed on top of the box and slapped it before eventually hitting the touchscreen and earning his reward. “It was almost like you could see him realizing, ‘Ah, that’s what you have to do, touch the screen!'” Sánchez Vargas said in a statement. “It was amazing to watch an individual learn something so quickly.”

Over six sessions with two habituated groups, 16 capuchins interacted with the box while another 14 watched from a distance without ever touching it. Of the 16 that engaged, 10 learned to reliably trigger a reward, and eight built what the researchers called “robust screen-reward associations” — meaning they kept coming back to touch the screen and expect a banana. Five of those monkeys still remembered the habit a full week after the device was removed from the reserve. Not every monkey used the same technique, either; some tapped the screen with their hands, while others learned to touch it with their lips.

Why It Matters Beyond the Jungle

Study co-author Marcela Benítez, an assistant professor of anthropology at Emory, said the appeal of the system is that it lets scientists study primate decision-making in a natural setting rather than a controlled lab. “The primate brain didn’t evolve in a lab, it evolved in complex, competitive environments,” Benítez said. “Yet primate cognition is rarely studied in the wild because the experimental control needed to measure cognition is difficult in unpredictable environments.” That distinction matters for anyone studying gambling-adjacent behavior generally, since a controlled lab setting can smooth over the kind of noisy, competitive social pressure that shapes real decisions — whether among wild primates fighting over a food dispenser or humans weighing a bet with friends watching.

The research team is already building on the prototype. They’re training an upgraded facial-recognition model to identify all 16 pilot monkeys individually so the device can serve up a task calibrated to each animal’s progress — testing categories like learning speed, impulse control, flexibility when the rules change, and short- and long-term memory. The researchers also plan to build in safeguards that limit how many rewards a single monkey can rack up per session, so one banana-hungry capuchin can’t just monopolize the machine and skew the data. It’s a reminder that even outside of a Vegas floor or a mobile sweepstakes app, reward loops built around anticipation and variable payout are powerful enough to reshape behavior fast — whether the player is human or not.